SIAM Industry Panel Offers Firsthand Look at Careers in Data Science and Machine Learning

Exciting new opportunities in machine learning and data science are drawing a large number of applied mathematicians and computational scientists to industry. In August 2023, the SIAM Industry Committee organized a virtual panel in which researchers from different industrial sectors shared their experiences outside of the academic sphere. Panelists Amr El-Bakry (Exxon), Ben Fogelson (Recursion Pharmaceuticals), Genetha Gray (Edward Jones), Nandi Leslie (Raytheon), and Emmy Smith (Amazon) offered insights based on their individual career paths, while moderators Christine Harvey (Mitre Corporation) and Nessy Tania (Pfizer) fielded audience questions about the broad and rewarding pathways that are available in industry.

El-Bakry opened the session by acknowledging that he worked in academia for a few years after earning his Ph.D. before trying an industry position on the advice of a friend. He thought that it would be a short-term venture; instead, he ended up staying. “What made it a long-term career is that I have always had a fascination with math and statistics technology,” El-Bakry said. “The second thing [that attracted me to industry] was seeing a day-to-day impact, including making money for the company where I work.”

Several other panelists also disclosed that they had originally planned to pursue careers in academia, but were attracted to industry for a multitude of reasons. “I had every intention of being a math professor,” Smith said. “Then I took statistics and kind of fell in love with how applicable it was and everything you could do with it.” Fogelson similarly intended to become a professor of mathematical biology, but he was unenthused by the lack of control that academics often have over the course and location of their career. However, he had initial concerns about industry as well. “Like a lot of people coming out of academia, I was nervous about industry and worried that it would be boring and repetitive,” Fogelson said. But on the contrary, he stated that his industry experience has involved fruitful collaborations with people from a wide variety of backgrounds to develop rapid impacts for interesting problems — with the added bonus of higher pay and an easier work-life balance than academia.

When discussion turned to the most relevant educational components for an industrial career, Smith touted so-called “soft skills”—such as communication, active listening, and the ability to respond to constructive criticism—as critical for success in any work environment. “A lot of times, I do work with folks for whom you put numbers in front of them and they glaze over,” she said. “I had to learn a lot along the way; I had a lot of missteps and good lessons learned.” The ability to see data at a granular level, then pull back to understand the entire context of a business and tell a compelling story, is essential.

“I’d echo the soft skills,” Fogelson said. “That was something I quickly learned when I made the switch to industry.” He also recommended that listeners practice the art of closely reading technical papers. “One of my superpowers at work is being able to read something and quickly learn a new practice,” Fogelson continued. “Bringing that value is a big part of how I make an impact.” Academic courses on optimization, statistics, mathematical logic, and programming are all likewise useful, as is the ability to read and write code documentation.

An attendee then asked about common misconceptions that students might have regarding machine learning in industrial settings. “One is that the data is clean and ready for analysis, and two is that the algorithm will work right away the first time we try it,” El-Bakry said. The panelists all agreed that contrary to popular belief, companies do not always have perfect setups with the newest technologies. Additionally, the process of determining whether systems are working correctly is often murky and costly, and data for certain questions might not exist or be legally accessible.

“Another misconception is that you can bring your full innovative and creative self to every problem,” Leslie said. “Sometimes you have to use a canned algorithm that has already been tested.” Smith concurred, noting that researchers are often eager to experiment with fun, innovative techniques that are unnecessarily complex for the question at hand. She advised attendees to maintain simplicity in order to avoid the trap of over-innovation.

Next, an audience member inquired about the significance of publication in industry. “One of the misconceptions of industry is that it’s one wholistic thing,” El-Bakry replied. “We have many different sectors, and the culture might be different [in each].” Businesses emphasize the factors that bring value to them, so winning contracts or contributing to successful proposals may be more beneficial than publishing papers; moreover, the level of importance of these activities at one company can even change over time.

Leslie offered her own take on the publications query. “I’d like to add patents and other intellectual property to that question,” she said. “Some companies may not care as much about publications in a journal, but they might care about patents.” The relevance of publishing also varies based on whether one is pursuing a management or research track.

Conversation then shifted to the topic of qualifications, as it is sometimes difficult to discern the appropriate level of formal education for industry positions. “In your interview and at certain companies, a Ph.D. is going to open a lot of doors,” Gray said. However, people with Ph.D.s do not automatically receive senior roles or titles; in some cases, their managers may not have doctorates but might nonetheless possess far more experience and knowledge about the business. Gray advised listeners to find their own spaces and refrain from comparing their stations with those of their colleagues, especially since Ph.D.s do not affect industry promotions in the same way as academia.

Furthermore, some industry-based machine learning and data science roles do not require Ph.D.s at all. A master’s degree, however, does often allow for more interesting research possibilities than a bachelor’s degree, while a Ph.D. is generally required for positions that actively shape the future direction of an organization. Job seekers should carefully review position listings to judge the requirements and discern the responsibilities beyond a potentially ambiguous title. “Take a look at different job descriptions and what they actually ask for,” Smith said. “Companies don’t always name what they want, and it’s important to read those job descriptions and ask questions.”

Because industry resumes are written and formatted differently than academic CVs, prospective applicants might want to solicit feedback from an acquaintance who already works in industry. Specific details about the impact of one’s work are essential; even if candidates do not have much actual work experience, they can describe their involvement in a hackathon or other substantial project based on its hypothetical outcome in a real-world setting. This emphasis on impact helps candidates stand out against the many other job seekers who likely have similar credentials.

When interviewing potential employees, Smith looks for individuals who know about the industry, have done their research on the company, and can articulate their reasons for wanting to work there. Gray observed that most people who flounder during the interview process do so due to their lack of presentation skills; as such, she emphasized the importance of talking about impacts in a way that would appeal to a director or human resources leader. Fogelson, on the other hand, finds that many interviewees drop out in the coding stage due to gaps in their ability to translate their abstract knowledge into the creation of something useful.

Professional conferences, university career events, and SIAM career fairs all offer opportunities for interested individuals to learn more about the job market for machine learning and data science, and begin to make valuable connections. “My number one thing is your network,” Gray said. “Make sure your network is strong, and make sure that you’re connecting with people who do what you want to do.” In fact, having an acquaintance at a company increases the likelihood that one’s application will be viewed and progress to the next stage. “You have your colleagues from academia, but I suggest that you also go to conferences, meetings, and workshops and engage through talking to the people you meet — folks at your level or above,” Leslie said. “Exchange information in order to continue to build your network.”

To round out the hourlong session, the panelists emphasized that robust mathematical training provides an abundance of translatable skills to the data science and machine learning space. By keeping an open mind, applied mathematicians can encounter new and unexpected challenges from all corners of industry.